Rahul Simha department of computer science George Washington - - PowerPoint PPT Presentation

department of computer science

George Washington University Office hours: 1pm Tuesdays, 5pm Wednesdays

Big Dating:

Computer Science and Relationships

Rahul Simha

some stats Usage of online da-ng websites: (total registra=ons)

• OKCupid: 5 million
• Chemistry: 11 million
• E-harmony: 33 million
• POF: 40 million
• Tinder: 50 million
• Match: 96 million
• Badoo: 200 million

your data Exercise 1 Part 1: On the worksheet, write three things about

yourself (without iden=fying yourself) that will help you stand out in an online da=ng site

Part 2: Write down your height in inches but add a fudge

factor of +10 or -10 using the following rule: if you were born in an even-numbered month, add 10. Otherwise subtract 10. Thus, if your height is 67 inches and you were born in March, you would write 57

Follow remaining “snowball” instruc=ons

more stats Percep-on:

• 60% of US adults: “online da=ng is a good way

to meet people”

• 2/3 of online daters have gone on a date with

someone they met online

• 27% between ages 18-24 have used online

da=ng But … 5% of those in commi[ed rela=onships say they met online

why the gap? Exercise 2: At your table, come up with three reasons why

• nline matchups may not lead to commi[ed

rela=onships

Algorithmic matching:

• OKCupid
• EHarmony

Basic ideas:

• Ask lots of ques=ons
• Perform some kind of scoring and matching

computer science to the rescue

High dimensional data Large data size Privacy, security (height example) Algorithm design

computer science issues

Exercise 3: draw the points (1,2), (2,1), (2,2), (7,6), (8,7), (9,10), (1,9) on paper. How many clusters do they fall into? Exercise 4: how many dimensions are present in the survey data you filled?

high dimensional data

Exercise 3: draw the points (1,2), (2,1), (2,2), (7,6), (8,7), (9,10), (1,9) on paper. How many clusters do they fall into? Exercise 4: how many dimensions are present in the survey data you filled? Algorithmic challenge: effec=ve clustering of high dimensional data E-Harmony: 29 dimensions

high dimensional data

Distance measure:

• Given any two users, compute “how

compa=ble they are” Sort:

• Sort all users by compa=bility

For every user we now have a sorted list of other users, in order of preference

the scoring problem

Suppose we need to match people, e.g., H1 H2 H3 with R1 R2 R3 Example matching: H1 H2 H3 R1 R2 R3

the matching problem

Suppose we know “spousal” preferences:

H1 `s preferences: R2 R1 R3 H2 `s preferences: R2 R3 R1 H3 `s preferences: R2 R1 R3 R1 `s preferences: H1 H2 H3 R2 `s preferences: H3 H1 H2 R3 `s preferences: H2 H1 H3

So H1 would prefer R2 as spouse to R1 and R1 over R3

the matching problem

Consider

H1 `s preferences: R2 R1 R3 H2 `s preferences: R2 R3 R1 H3 `s preferences: R2 R1 R3 R1 `s preferences: H1 H2 H3 R2 `s preferences: H3 H1 H2 R3 `s preferences: H2 H1 H3

the matching problem

Example matching: H1 H2 H3 R1 R2 R3 Exercise 5: what is the problem with this matching?

Consider

H1 `s preferences: R2 R1 R3 H2 `s preferences: R2 R3 R1 H3 `s preferences: R2 R1 R3 R1 `s preferences: H1 H2 H3 R2 `s preferences: H3 H1 H2 R3 `s preferences: H2 H1 H3

the matching problem

Example matching: H1 H2 H3 R1 R2 R3 H3 and R2 will elope! Algorithmic challenge: devise an algorithm to create a stable matching

• 1. Ini=ally place all H’s in unmarried-list
• 2. while unmarried-list is not empty
• 3. Hi = lowest numbered from list
• 4. try R’s in order of Hi’s preference
• 5. if Rj is not matched, match Hi and Rj
• 6. else if Rj prefers Hi to current match then
• 7. match Rj with Hi
• 8. return current match to unmarried list

Can prove: provides a stable matching Exercise 6: do the H’s or R’s get the best deal?

the proposal algorithm

Demo Prac-cal applica-ons?

the proposal algorithm

Demo Prac-cal applica-ons:

• Med school internships
• Clerkships with judges

the proposal algorithm

Exercise 7: Go to h[ps://oracleomacon.org/ and enter actors in two movies YOU have seen. Try to find two actors with a distance of 4.

the (network) structure of rela-onships

The Milgram experiment Demo

the (network) structure of rela-onships

The Milgram experiment Demo Facebook: 4.57 (among 1.5b users) Da=ng app based on “who knows who”: Hinge

the (network) structure of rela-onships

Exercise 8: Choose between

• 1. Go to h[p://www.masswerk.at/elizabot/ and

converse with Eliza.

• 2. Talk to Siri and record the exchange on paper.
• 3. Volunteer as judge.

Compe-tors: your conversa=on must be short (two back-and-forths) Judges: pick the best conversa=on. Remember Robert Epstein?

talk to a bot

Exercise 9: Choose between

• 1. Yes, it’s fine for humans to marry robots in

the future.

• 2. No, that should never be allowed.

Write down your reasons on the worksheet

• ur future with robots

summary

Computer science under the hood:

• Programming of websites (for online da=ng)
• Servers, networks, clouds, large data:

Ø Example: 25 TB of data at E-Harmony, incl. 200+ million images

• Algorithms for matching, graph structure
• Algorithms for clustering, machine learning,

natural languages

• Robo=cs / AI